Intel® oneAPI Deep Neural Network Developer Guide and Reference
A newer version of this document is available. Customers should click here to go to the newest version.
Visible to Intel only — GUID: GUID-96837589-24DD-4BC0-A849-7F21BFD8D604
Visible to Intel only — GUID: GUID-96837589-24DD-4BC0-A849-7F21BFD8D604
enum dnnl_graph_partition_policy_t
Overview
Policy specifications for partitioning. More…
#include <dnnl_graph_types.h> enum dnnl_graph_partition_policy_t { dnnl_graph_partition_policy_fusion = 1, dnnl_graph_partition_policy_debug = 2, };
Detailed Documentation
Policy specifications for partitioning.
Enum Values
dnnl_graph_partition_policy_fusion
Fusion policy returns partitions with typical post-op fusions, eg.
Convolution + ReLU or other element-wise operations or a chian of post-ops.
dnnl_graph_partition_policy_debug
Debug policy doesn’t not apply any fusions.
It returns partitions with single operation in each partition. The policy is useful when users notice any bug or correctness issue in fusion policy.